AMD Instinct MI325X product page 'Specs Comparisons' charts - NVIDIA H200 figures printed WITHOUT a sparsity label
Fields
sources/vb-amd-mi325x-h200-comparison-chart-dense-basisNotes
A GENUINE TWO-SIDED VENDOR DISAGREEMENT, verified verbatim. AMD's charts against H200 SXM print 'TF32 494.7 | 653.7', 'FP16/BF16 (Tensor/Matrix) 989.4 | 1307.4', 'FP8 1978.9 | 2614.9', 'FP64 (Vector) 33.5 | 81.7', 'Memory Capacity 141 GB | 256 GB', 'Memory Bandwidth 4.8 TB/s | 6 TB/s' - with NO sparsity label on the H200 column, unlike the MI300X page which DOES carry '(Sparsity)'. AMD's own footnote on this page states H200 at '989.5 TFLOPS peak theoretical half precision tensor (FP16 Tensor)'. So AMD prints the H200's SPARSE figure while labelling it as if dense, whereas NVIDIA's own materials footnote the same magnitude as 'with sparsity'. The MI300X page labelling its comparison chart and this page NOT labelling it is itself inconsistent AMD behaviour. Both are primary vendor statements; the disagreement is a property of the sources. No effect on the MI325X record's own rows.
Cross-references
This record references no other records.